Short Communication: Zipf-Aware Token Budgeting for LLM Context Windows
Abstract
We propose a Zipf-aware policy to trim low-utility spans before retrieval, lowering context cost 22% with negligible answer loss.
Cite this article
Okafor, A. & Roth, D. (2025). Short Communication: Zipf-Aware Token Budgeting for LLM Context Windows. Research Explorations in Global Knowledge & Technology (REGKT), 3 (8). Retrieved from https://regkt.com/article.php?id=151&slug=short-communication-zipf-aware-token-budgeting-llm-context-windows